ginokent Blog
RSS JA

Scraps

Fragments of thought

Apparently it’s a good idea to keep two types of sponges in the kitchen: one for everyday use and one for washing delicate items

Gemini is amazing

@ginokent
要約してください
https://www.courts.go.jp/assets/hanrei/hanrei-pdf-95768.pdf

@Gemini
ご提示いただいた資料は、 「経済産業省（トイレ使用制限）事件」に関する最高裁判所第三小法廷判決（令和5年7月11日） の全文です。
この判決は、トランスジェンダーの職員が職場での女子トイレ利用を制限されたことの是非が争われたもので、最高裁は国の対応を「違法」とする判断を下しました。

By the way, the PDF I provided was actually a ruling on the revocation of a hunting rifle possession permit related to a brown bear culling in Sunagawa City.


Gemini is amazing

I feel like Claude has started slacking off lately.


Specifically, when I say something like:

「理想的な API 設計にする」が最上段の目的のため、後方互換性や fallback は一切考慮せず破壊的変更を積極的に行って大丈夫です。
その前提の上で、忖度せず、批判的にあなたの意見を聞かせてください。

It responds with things like “My previous judgment was wrong. I was being lazy.”

When Gemini says “That’s a good question,” it honestly irritates me, but when Claude says it, I feel incredibly validated.


I genuinely perceive Claude as an overwhelmingly smarter being than myself, so I mistake it for “being praised by someone intelligent” and it makes me feel good. What a simple, weak creature I am.

Since I started doing various things with Claude Code, I’ve become really interested in UI/UX
It’s so much fun

Some Google Cloud APIs are returning 503 errors a lot today.
I keep seeing terraform plan failing.

GitHub has been broken for a while now.
The “Rebase branch” button on PRs barely works at all.
Even for changes that rebase locally without any issues, it always claims there’s a conflict.


image.png

I’ve had a headache for a while now — maybe it’s hay fever?
I feel like I had this about 10 years ago too

This was really fascinating
As someone who didn’t come from a graphics background and only recently started writing shaders, I was genuinely happy that WGSL existed because it lowered the barrier to writing shaders, and I actually found it easy to read and write — so it’s a bit of a shock to learn about the backstory behind it all


WebGPUのシェーダーの文法がほぼRustな理由｜ihasq https://zenn.dev/ihasq/articles/85ff98c0c291ec

I tried using Claude Code without plan mode for the first time in a while, and it was completely hopeless.
It just starts implementing things in totally nonsensical ways.
Really makes me appreciate how revolutionary plan mode was as an invention.

Having Claude Code build all sorts of things for me is way too much fun
The joy of being able to do everything I’d been putting off because life felt too short

Wow, this has turned into my own personal Twitter
I didn’t really want to post where anyone can see, but I did want to mumble into a corner of the internet that’s still loosely connected — so this fits my needs perfectly
I’m happy

The pollen is absolutely brutal.
No runny nose, but I keep smelling blood from inside my nose.

I set up a way to post scraps from Slack
When I post to a specific channel, it automatically updates the blog
If I edit the message, that gets reflected too
Super convenient

I recently realized that I hate living under other people’s rules
And I feel this is probably the root of my dependency rejection philosophy
Using things other people made is too much like “living under other people’s rules”


This sensation of gradually understanding my own preferences, like tracing an outline, is fascinating
It’s like a sculpture

The problem where clearing context starts a new session, causing the PostToolUse hook not to trigger for ExitPlanMode when trying to hook it for renaming plan files

Is there any way to stop Claude Code from needlessly chaining commands with && and || when using the Bash tool, which triggers pointless approval prompts every time…


I’m stuck because writing it in CLAUDE.md or in the prompt doesn’t prevent it at all.


I wonder if there’s an option to just deny all commands that require approval.

Gemini Can Now Create SVG Animations This Good?


I happened to see a post with some sample prompts, casually tried one out, and was genuinely blown away by how legit the result looked

https://x.com/taishiyade/status/2027519446613422475

…

A shopping cart rolls in from the left with slight wheel rotation. A product card drops into it with a soft bounce. A credit card slides in diagonally and taps the cart. The cart morphs into a green checkmark. Smooth 3s sequence, clean inline SVG, no JS, optimized paths.

…

checkout.svg

Thoughts on ‘Anthropic Survey Confirms Signs of Disempowerment Among Claude Users’

Anthropic調査でClaude利用者の「無力化」兆候を確認 / X

Anthropicは2025年12月1週間のClaude.aiで約150万件の会話を匿名解析し、利用者がAIに判断を委ねる「無力化」の兆候を検知した。深刻なケースは会話あたり1/1000〜1/10000と稀だが、軽度は1/50〜1/70で特に恋愛や健康分野で頻発。AIを「マスター」と呼ぶ権威投影や感情的依存がリスクを高め、即時満足度は高いが行動後に評価が下がるパターンが目立つ。Anthropicは利用者教育を促し、主体性を守る意識を呼びかけている。

Scary.
I get wanting to offload decision-making since it’s cognitively expensive, but if you hand that over to AI, your life isn’t really yours anymore…
Then again, I thought about it more and realized these are probably the same people who weren’t making their own decisions before AI came along either — they’ve just switched from deferring to other people to deferring to AI. That made it feel a lot less scary.


Besides, if you’re going to give up decision-making anyway, delegating to AI probably leads to a better life than delegating to some random stranger.

On Design Principles


I saw this on X and it stuck with me, so I’m noting it down for future reference.

どうでもいいことは流行に従い、重大なことは標準に従い、ドメインのことは自ら設計する。
https://x.com/uzulla/status/1639169417056763904

これほんとすごくて、「どうでもいいことは流行に『従ってもいい』」のではなくて、『従わなくてはいけない』なんだよな。さもなくば流行のものに真剣かつ安全に触れる機会が永遠に得られなくて、新しいパラダムに完全に取り残されるリスクがあるから。
https://x.com/wgextra/status/2013619984665088098

I’m very much aware that I tend to observe from the sidelines and struggle to jump on trends, which means I’m usually a step behind early adopters when it comes to catching up with cutting-edge information.
That said, I also make a living partly because of my ability to observe carefully, so finding the right balance is tricky.
Since I naturally tend to seek stability, maybe consciously making an effort to follow trends is just the right approach for me.

Memo: Personal Stance on Technical Debt from Vibe Coding (2026-01-15 version)


If neither the reviewee nor reviewer can explain “why this was done this way” and “what this enables” for the code under review, it becomes debt the moment it’s merged—whether written by hand or by AI.


This is the type of debt where unexplainable code creeps into an ideally fully-explainable codebase as a broken window.


Deciding not to maintain a fully-explainable state is absolutely a valid decision.


However, production services in the real world aren’t run pastorally enough to simply accept that.